This website uses cookies

Read our Privacy policy and Terms of use for more information.

// AI Lessons

Read an AI Vendor's 10-K in 20 Minutes

Pull any public vendor's 10-K and let a chatbot surface the three signals that predict whether they survive the AI stack.

So one line from my dot-com years has aged better than almost anything else I picked up: the worst business to be in is a commodity business where you're not the 800-pound gorilla. Everybody else just pushes your margins out until there's nothing left. I watched a dozen companies in the early 2000s that were really just a thin layer of software sitting on top of somebody else's real product. The platform underneath added their feature as a checkbox, and they were gone. Not acquired. Gone.

I'm watching the same movie again. Except now the "platform underneath" is a frontier AI lab, and the thin layer is half the software you're about to renew a three-year contract on.

Here's the uncomfortable part: you can usually tell which vendors are durable and which ones are about to get eaten. The information is public, it's free, and it sits in a document called a 10-K that almost nobody outside of finance actually reads. Until this year, getting through one took training or a lost afternoon. Now it takes a chatbot and about 20 minutes.

Read the vendor's own filing before you renew, not their sales deck.

// The Takeaway: Before you sign or renew a multi-year deal with any public AI or software vendor, pull their latest 10-K or quarterly filing (free, from SEC EDGAR) and run it through any chatbot with the prompt below. In 20 minutes you surface three signals: is revenue growing or shrinking and how dependent are they on a handful of customers; is their gross margin healthy or are they just reselling someone else's model; and do their own risk factors admit AI could disrupt them, with enough cash to survive it. That tells you whether you're buying from a durable "shovel-seller" or a "disruption target" wedged between the frontier labs and you. This is vendor due diligence, not investment advice.

Open SEC EDGAR full-text search and pull your vendor's 10-K → About 20 minutes. Free tier of ChatGPT, Claude, or Gemini works — paste the filing text, or use any model with web or file access. Verify every number against the filing; chatbots do hallucinate figures.

This week the theme is following the money: what the market already priced in that your roadmap hasn't. Monday reported it. Today you learn to read it yourself, one vendor at a time.

// The real shift: For twenty years, "is this vendor any good?" meant checking references and the product roadmap. That question is now downstream of a bigger one: does the AI stack make this vendor's whole category cheaper to build than to buy? A company can have a great product and a happy customer list and still be a disruption target, because the frontier labs just turned its core feature into a $20-a-month commodity. The filing is where a company is legally required to tell you that — in the risk factors, in the margin line, in whose model it actually runs on.

The three signals, and where they hide

You're not doing accounting. You're reading three things a vendor has to disclose and would rather you skimmed past.

Signal 1 — Revenue trend and customer concentration. Find total revenue this year versus last, and the sentence about customer concentration. Growing revenue with a diversified base means durable. Shrinking revenue, or language like "a limited number of customers account for a substantial portion of our revenue," means fragile — you'd be one renewal in a stack of renewals holding the whole company up.

Worked example: C3.ai. Its fiscal-2026 results show revenue fell to $250.3 million from $389.1 million a year earlier — a 35.7% drop, and the 10-K states plainly that "a limited number of customers have accounted for a substantial portion of our revenue". That's the profile of a disruption target, and the stock spent the year confirming it.

Signal 2 — Gross margin and whose model they run. Gross margin tells you whether they own something scarce or resell something abundant. Real shovel-sellers — infrastructure, proprietary data, actual IP — hold fat margins. A thin wrapper on someone else's model runs thinner, and the tell is right there in the text. Search the filing for "LLM," "third-party," and "open source." C3.ai's own 10-K describes its platform as "LLM-agnostic", supporting both commercial and open source models. That can be a smart strategy — but it means the scarce asset lives upstream, at the labs. And its GAAP gross margin sat at 31%, far below the 70%-plus that healthy enterprise software throws off.

Signal 3 — Disclosed AI risk and cash runway. Companies must disclose what could kill them. Search the risk-factors section for "artificial intelligence," "competition," and "technological change." C3.ai's filing warns it faces "intense competition" and that if it fails "to respond to rapid technological changes... our ability to remain competitive could be impaired." Then check cash. C3 held about $673 million against a non-GAAP annual loss near $190 million — roughly three years of runway. So: shrinking top line and thin margins (fragile), but not about to miss payroll (survivable). The three signals don't always point the same way, and that split is the whole point — you're grading durability, not writing an obituary.

The 20-minute move

  1. Get the filing, free. (5 min) Open EDGAR full-text search, type the company name or ticker, filter the form type to 10-K (annual) or 10-Q (quarterly), and open the latest one.

  2. Feed it to a chatbot. (2 min) Paste the text — free tier is fine — or hand a model with web or file access the ticker or the EDGAR link.

  3. Run the prompt below. (3 min) Copy it whole.

  4. Verify the three numbers. (5 min) Ctrl-F each figure it cites in the actual filing. If a number isn't there, don't trust it. This is the hallucination check, and it's not optional.

  5. Read the risk factors yourself. (5 min) Let the chatbot point you to the section, then read the AI and competition language in the company's own words.

  6. Score it. Shovel-seller, disruption target, or mixed — and how many years of runway.

Here's the prompt. Copy it whole, then paste the filing or drop in the ticker:

You are a skeptical equity research analyst doing vendor due diligence for a
buyer. This is NOT investment advice. I will give you a public company's
latest SEC filing (10-K or 10-Q), or its name/ticker if you can retrieve the
filing yourself. Work ONLY from the filing. If a number is not in the filing,
say "not disclosed" - never estimate or fill it in from memory.

[PASTE THE 10-K / 10-Q TEXT HERE, or give the company name/ticker]

Return exactly three signals. For each, give the figure AND quote the sentence
you pulled it from:

1. REVENUE & DEPENDENCY. Total revenue this period vs. the prior period and the
   percent change. Quote any sentence about customer concentration (a few
   customers making up a large share of revenue). Verdict: growing/diversified
   or shrinking/concentrated.

2. MARGIN & WHOSE MODEL. Gross margin (or gross profit divided by revenue).
   Then search for how they source their AI: quote any sentence mentioning
   third-party, commercial, or open source LLMs/models. Verdict: owns a scarce
   asset (healthy margin, proprietary tech) or resells someone else's model
   (thinner margin, model-agnostic).

3. AI-DISRUPTION RISK & RUNWAY. Quote the risk-factor sentences mentioning
   artificial intelligence, competition, or rapid technological change. Then
   give cash and equivalents on hand and the annual net or operating loss, and
   estimate years of runway (cash divided by annual loss).

End with: a one-line verdict - "durable shovel-seller," "disruption target,"
or "mixed" - and a list titled "VERIFY THESE NUMBERS AGAINST THE FILING" with
every figure you used, so I can Ctrl-F each one myself.

One honest caveat, because it matters: the number-checking step is not decoration. Language models are excellent at reading a filing and genuinely bad at arithmetic — and worse, they'll produce a confident figure when they don't actually have one. The prompt tells the model to say "not disclosed" and to hand you a verification checklist. Use it. The filing is the source of truth; the chatbot is just the fastest reader you've ever hired.

Bonus: run it on your own renewal list

Pull the three vendor renewals landing in the next two quarters and run all three. You'll find at least one where the vendor's own filing says something their account exec never will.

Why this week

The AI market spent the last two years repricing the exact software categories most companies are mid-contract on — the seat-based tools that quietly got cheaper to replace than to renew. That repricing is a free signal, and now you can read it yourself before the renewal call instead of after it. Tomorrow's AI Advantage takes the next step: what to actually do at that call once you know which side of the line your vendor is on.

But anyhow — start with one filing. The vendor whose contract you'd have to sign next.

Your AI Sherpa,

Mark R. Hinkle
Founding Publisher, The AIE Network
Follow me on LinkedIn

If you want to get in contact or give me feedback, reply to this email. I read every single one of them.

Keep Reading